Fitting the Cox proportional hazards model to big data

Area of Science:

• Biostatistics
• Survival Analysis
• Computational Statistics

Background:

• The Cox proportional hazards model is standard for analyzing time-to-event data with covariates.
• Analyzing large datasets (big data) with traditional Cox model fitting methods is computationally intensive.
• Handling time-dependent covariates and censored data are key challenges in survival analysis.

Purpose of the Study:

• To propose a computationally efficient method for fitting the Cox proportional hazards model to big data.
• To reduce the computational burden of Cox model estimation for large-scale studies.
• To ensure the proposed method maintains the statistical properties of the conventional estimator.

Main Methods:

• Maximum partial likelihood estimation on a data subset.
• One-step estimation using efficient score functions to incorporate remaining data.
• Validation through extensive simulation studies and real-world data application (UK Biobank).

Main Results:

• The proposed method achieves the same asymptotic distribution as the full dataset estimator.
• Significant reduction in computation time compared to conventional methods.
• Demonstrated effectiveness and efficiency on large-scale UK Biobank data.

Conclusions:

• The proposed method offers a computationally efficient alternative for Cox model fitting in big data settings.
• This approach enables accurate survival analysis on massive datasets, previously limited by computational resources.
• The method is practical for large cohort studies and real-world data analysis.